Qdrant, a provider of composable vector search and database solutions, has rolled out upgraded indexing performance, triple-zone cluster failover mechanisms, and compliance-focused audit logging features.
The firm develops a standalone vector database for storing embeddings required by large language models (LLMs) and autonomous AI agents, which commonly power AI inference workflows such as Retrieval-Augmented Generation (RAG). According to Qdrant, enterprise procurement teams always evaluate vector search tools against three core criteria: scalability for growing workloads, service continuity amid infrastructure failures, and traceable operational auditing.
André Zayarni, Co-Founder and CEO of Qdrant, commented: “GPUs are no longer limited to model inference; they also optimize data indexing. We have enabled GPU-accelerated HNSW construction in our open-source version since v1.13, and this capability is now officially available on Qdrant Cloud. Combined with multi-AZ replication and audit logging, the integrated suite equips enterprises to deploy Qdrant for critical production workloads.”
Qdrant’s latest product upgrades cover three key enhancements:
GPU-accelerated indexing: Benchmark tests confirm dedicated GPUs boost HNSW index construction speed by up to four times on Qdrant Cloud. Users can mount GPU resources to existing clusters to handle high-intensity indexing bursts efficiently.
Multi-Availability Zone (AZ) clusters: The cross-AZ replication mechanism duplicates data across three availability zones in a single region. It eliminates manual failover latency, ensuring uninterrupted read-write operations even if one availability zone suffers an outage.
Audit logging: The function records all API-based operations, including data queries, upserts, deletions, collection management and snapshot tasks. Each log entry adopts structured JSON format, marking user identities, API keys, timestamps, target collections and operation authorization status. When autonomous systems execute tasks based on retrieved data, the logs deliver clear audit trails for request sources, execution time and access legitimacy. Users can customize log retention cycles and export records externally via APIs for long-term archiving.
Currently, GPU-accelerated indexing is accessible on AWS, with ongoing plans to expand coverage to more cloud vendors and regions. Multi-AZ clusters belong to Qdrant’s Premium tier, delivering an SLA-backed uptime of 99.95%. The audit logging feature is open to all paid Qdrant Cloud clusters.
Official documentation provides further details on the three new capabilities.
Footnote
Hierarchical Navigable Small World (HNSW) is an algorithm designed to identify vector nearest neighbors. It maps vectors into interconnected graph structures, which expand drastically with growing data volume. HNSW stacks multiple virtual graph layers: the top sparse layer contains minimal vectors for rapid preliminary searching, while lower layers store increasingly more vectors until the bottom layer covers all data points. Each layer acts as an optimized entry point for the next, greatly shortening overall search latency.
Qdrant faces competition from multiple industry peers. Pinecone also adopts HNSW algorithms and leverages NVIDIA GPUs to optimize embedding and reranking performance. Zilliz delivers both HNSW compatibility and GPU acceleration; its underlying Milvus database integrates Nvidia CUDA-Accelerated Graph Index for Vector Retrieval (CAGRA) from the RAPIDS cuVS library to enable GPU indexing.
Beijing Qianxing Jietong Technology Co., Ltd.
Sandy Yang/Global Strategy Director
WhatsApp / WeChat: +86 13426366826
Email: yangyd@qianxingdata.com
Website: www.qianxingdata.com/www.storagesserver.com
Business Focus:
ICT Product Distribution/System Integration & Services/Infrastructure Solutions
With 20+ years of IT distribution experience, we partner with leading global brands to deliver reliable products and professional services.
“Using Technology to Build an Intelligent World”Your Trusted ICT Product Service Provider!
Your message must be between 20-3,000 characters!
English
